Research Articles: Behavioral/Cognitive Task related sensorimotor adjustments increase the sensory range in electrolocation https://doi.org/10.1523/JNEUROSCI.1024-19.2019
Cite as: J. Neurosci 2019; 10.1523/JNEUROSCI.1024-19.2019 Received: 14 May 2019 Revised: 9 November 2019 Accepted: 18 November 2019
This Early Release article has been peer-reviewed and accepted, but has not been through the composition and copyediting processes. The final version may differ slightly in style or formatting and will contain links to any extended data.
Alerts: Sign up at www.jneurosci.org/alerts to receive customized email alerts when the fully formatted version of this article is published.
Copyright © 2019 the authors
Title: Task related sensorimotor adjustments increase the sensory range in electrolocation
Abbreviated title: Sensorimotor learning improves the sensory range
Authors: Federico Pedraja1, Volker Hofmann1,2, Julie Goulet1, Jacob Engelmann1*
Affiliations: 1 Bielefeld University, Faculty of Biology/CITEC, AG Active Sensing, Postfach 100131, D-33501 Biele- feld, GERMANY 2 McGill University, Department of Physiology, 3655 Promenade Sir William Osler, Montreal, QC, H3G 1Y6 CANADA * Corresponding author: Jacob Engelmann AG Active Sensing Bielefeld University D-33501 Bielefeld, Germany Tel +49-521-106-4641 [email protected] Number of pages: 38 Number of figures: 9 Number of words: abstract (163 words), introduction (556 words), discussion (1609 words).
Competing interests: Authors of this manuscript do not have any financial or non-financial compet- ing interests. Acknowledgments: This work was supported by the Cluster of Excellence Cognitive Interaction Tech- nology ‘CITEC’ (EXC 277) and the DFG (EN 826/5-1).
ABSTRACT
1 Perception and motor control traditionally are studied separately. However, motor activity can serve 2 as a scaffold to shape the sensory flow. This tight link between motor actions and sensing is particu- 3 larly evident in active sensory systems. Here, we investigate how the weakly electric mormyrid fish 4 Gnathonemus petersii of undetermined sex structure their sensing and motor behavior while learn- 5 ing a perceptual task. We find systematic adjustments of the motor behavior that correlate with an 6 increased performance. Using a model to compute the electrosensory input, we show that these 7 behavioural adjustments improve the sensory input. As we find low neuronal detection thresholds at 8 the level of medullary electrosensory neurons, it seems that the behavior-driven improvements of 9 the sensory input are highly suitable to overcome the sensory limitations, thereby increasing the 10 sensory range. Our results show that motor control is an active component of sensory learning, 11 demonstrating that a detailed understanding of contribution of motor actions to sensing is needed to 12 understand even seemingly simple behaviors.
1
13 SIGNIFICANCE STATEMENT
14 Motor-guided sensation and perception are intertwined, with motor behavior serving as a scaffold to 15 shape the sensory input. We characterized how the weakly electric mormyrid fish G. petersii, as it 16 learns a perceptual task, restructures its sensorimotor behavior. We find that systematic adjustments 17 of the motor behavior correlate with increased performance and a shift of the animal’s sensory at- 18 tention. Analyzing the afferent electrosensory input shows that a significant gain in information re- 19 sults from these sensorimotor adjustments. Our results show that motor control can be an active 20 component of sensory learning. Researching the sensory corollaries of motor control thus can be 21 crucial to understand sensory sensation and perception under naturalistic conditions.
2
INTRODUCTION
22 Exploratory behavior is a crucial substrate for learning (Loewenstein, 1994). As the animals’ move- 23 ments influence the sensory input, re-organizing the motor patterns with respect to recent experi- 24 ence may contribute to learning thereby improving behavior. Analyzing these modifications can thus 25 reveal how motor action contributes to learning (Wolpert and Landy, 2012; O’Hora et al., 2013) and 26 aid decision making through action selection (Charlesworth et al., 2011; O’Hora et al., 2013; 27 Zgonnikov et al., 2017). 28 While the variability of motor behavior may facilitate motor learning by widening the search space 29 from which behaviors are instantiated (Brainard and Doupe, 2013; Wu et al., 2014), the same varia- 30 bility can set bounds on the task-optimization of motor control (van Beers et al., 2002). This is partic- 31 ularly evident in active sensory systems, where the sensory input directly depends on the motor out- 32 put. Here the strong sensorimotor dependencies may be exploited by an animal to adjust motor be- 33 havior in order to not only improve the motor but also the sensing efficiency (Friston, 2010; Little and 34 Sommer, 2013; Gordon et al., 2014). 35 36 We here investigated how sensorimotor behavior changes while Gnathonemus petersii, a pulse type 37 weakly electric fish, learned a detection task. During active electrolocation these fish obtain sensory 38 information through brief discharges of a specialized electric organ in their tail (electric organ dis- 39 charge, EOD). The discharge rate is under top-down control and changes in a context-dependent 40 manner (Post and von der Emde, 1999; Caputi et al., 2003). Each emitted EOD creates a 3- 41 dimensional electric field around the fish which is perturbed by nearby objects (Lissmann and 42 Machin, 1958). Also motion of the animal can perturb the electric field (e.g., tail movement (Sawtell 43 et al., 2006)), both of which are perceived by electroreceptors in the skin of the fish. To discriminate 44 between the predictable (re-afferent) and unpredictable (ex-afferent) components of the sensory 45 input, weakly electric fish are known to rely on a sophisticated neuronal circuitry (Sawtell et al., 46 2005; Bell et al., 2008) which enables them to analyze their nearby environment. 47 Not all (re-afferent) sensory consequences of behavior must be unfavorable however: Similar to oth- 48 er organisms (Poteser and Kral, 1995; Kern et al., 2001), weakly electric fish exhibit a variety of stere- 49 otyped behaviors (Toerring and Belbenoit, 1979; Toerring and Moller, 1984; Nelson and Maciver, 50 1999; Hofmann et al., 2014). Recent studies have revealed that behaviorally relevant sensory infor- 51 mation can emerge from such strongly patterned sensorimotor behaviors, i.e. weakly electric fish 52 actively exploit these sensorimotor dependencies (Hofmann et al., 2017; Pedraja et al., 2018).
3
53 The ability to control the timing of sensory sampling while at the same time being able to shape the 54 properties of the sensory input through their motor behavior makes weakly electric fish particularly 55 suitable to study how changes in exploratory behaviors can guide sensory-driven learning efficiently. 56 We here focussed on a reinforced object detection task and found that performance was progres- 57 sively enhanced by consistent changes of the motor patterns. These changes resulted in an increased 58 sensory range. Our results add further support to the idea that weakly electric fish actively improve 59 sensing capabilities by selecting purposeful components from their motor repertoire and focus their 60 electric attention in a goal-directed manner. Such behavioral control of the sensory input might con- 61 tribute to improving neuronal stimulus detection and encoding, as we found neuronal performance 62 to be relatively poor at the level of the medulla. 63
4
64 MATERIAL AND METHODS
65 Animals. Wild-caught Gnathonemus petersii of either sex were obtained from a commercial fish 66 dealer (Aquarium Glaser, Rodgau, Germany) and housed in communal 400L aquaria. The water tem- 67 perature in these aquaria and the set-up was 25 ± 1 °C at a conductivity of 100 ± 5 μS cm−1 and a 68 12L:12D photoperiod. Fish were fed with bloodworms. All procedures for animal maintenance and 69 preparations comply with the current animal protection law of the Federal Republic of Germany and 70 have been approved by the local authorities (Landesamt für Natur, Umwelt und Verbraucherschutz 71 Nordrhein-Westfalen: 87–51- 04.2010.A202).
72 Behavior
73 Training setup. Five G. petersii (length of 11 ± 1 cm) housed in separate experimental tanks and fed 74 with bloodworms to satiation three times per week before the beginning of the behavioral experi- 75 ment. The experimental tanks (120 · 50 · 50 cm) were divided in a living area (60 · 50 cm; 30 cm, wa- 76 ter level) that was separated by a plastic gate from the experimental area (60 · 50 cm; 10 cm, water 77 level). The floor in the experimental area was 20 cm above the floor of the living area, which con- 78 fined the movements of the animal in the experimental area into two dimensions. A plastic plate 79 divided the proximal end (20 cm) of the experimental arena in two compartments. Perpendicular to 80 this plate a 1 cm wide plastic strip marked the entry to the compartments on the floor. A metal cube 81 (2 · 2 · 2 cm) was placed on the floor at the decision line, centered in front of the cued compartment 82 where it served as a cue to the rewarded compartment (S+). Experiments were performed in dark- 83 ness (< 0.1 lux measured above the water level) and videotaped from the top (60 fps; AVT Marlin F- 84 131 & F-033) using IR-illumination (880 nm) from below. This wavelength is beyond the perceptual 85 range of this species (Ciali et al., 1997). EODs were recorded differentially (custom-built electrode 86 array, 0.6 – 40 kHz band pass) and stored as events (PC audio card, 12 bit, 10 kHz) alongside acquired 87 videos.
88 Experimental design, video tracking and statistical analysis. Animals first learned to swim through 89 the opened gate to the proximal end of the arena to receive food. Once fish did this reliably training 90 commenced, and videos of each trial were acquired. Each trial started by opening the gate and end- 91 ed with the fish swimming back to the living area. Crossing of the decision line was scored as a choice 92 and correct choices were food rewarded. After the fish returned to the living area the gate was 93 closed. Which of the two compartments was cued was determined in a pseudo-random fashion 94 (Gellermann, 1933). The cube was removed from the tank and re-positioned after each trial, even 95 when the same compartment was cued in consecutive sessions. Training was done for six days a 96 week with one session of 20 - 30 trials per day. When the performance reached 80% correct trials in 5
97 six consecutive sessions, learning we considered learning to be completed. We then separated the 98 data of each fish into three learning stages: Stage I contained the first six sessions where perfor- 99 mance was below 60% (511 trials from 5 animals); Stage II the consecutive six sessions (530 trials) 100 where performance was >60% and <80% and stage III comprises the first six session after the fish 101 exceeded 80% performance (592 trials). 102 In addition, we determined the sensory detection range after learning was completed by gradually 103 increasing the distances of the cue from the decision line (2 – 12 cm). At each distance, 20 - 40 test 104 trials were conducted before increasing the test distances. Similar to the training procedure, sessions 105 were done for six days a week with 20 control and 10 test trials per session. The detection limit was 106 determined as the object distance where the sigmoidal fit to the performance reached 75%.
107 Using a background subtraction approach the animals’ center of mass was determined off-line using 108 custom written MATLAB routines (R2016b 64 bit, MathWorks, Natick, MA USA). The posture of the 109 animal was obtained by applying a 3rd order polynomial fit through the midline of the body. Head and 110 tail positions were determined based on the spindle-like shape of the fish’s body with the head being 111 closer to the body’s center of mass. The position of the object was also tracked. From the change of 112 the animal’s position between consecutive frames we determined the 2D kinematics (i.e. thrust, slip 113 and yaw velocity) which were used for the behavioral classification (see Hofmann et al. (2014) for 114 more information). In trials where the cue was in the left compartment the data was mirrored along 115 the long axis of the arena in order to have movement prototypes in consistent relation to the cued 116 compartment.
117 To quantify the spatial distribution of behaviors the arena was binned (bin size 1 · 1 cm). The distance 118 to the cube was measured as the Euclidean distance between the fish’s head and the object. In trials 119 where the fish chose the wrong compartment, distance was calculated with respect to the virtual 120 object position (i.e., we assume the object to be present in the compartment the fish had erroneous- 121 ly chosen). To illustrate the average trajectories per fish and learning stage, we obtained the mean 122 direction in which fish passed from one spatial bin to the next and the corresponding vector strength 123 ( = ∙ with being the number of elements in the bin and the Rayleigh’s coefficient of 124 angular dispersion). Based on these values the average gradient of the trajectories was visualized 125 using the streamline-function in MATLAB. This only served as a visualization-tool, while all analysis is 126 based on single trajectories. Sampling density (SD) was calculated as the number of EODs emitted per 127 distance traveled (EOD count · cm-1) using swim speed and the EOD rate per frame. To calculate at- 128 tracting sets, spatial maps were generated from all trajectories. From each trajectory, the first coor- 129 dinate received a weight of +1 and the last coordinate of -1. Weights for coordinates in between 130 were linearly interpolated based on travelled time and distance. For each session, we superimposed
6
131 all weighted trajectories, resulting in cumulative 2D maps. The size of the attracting sets was defined 132 as the area in which data fell below a threshold which was defined by two standard deviations 133 (standard deviation was calculated based on all values for all fish) above the minimum value of the 134 map. 135 Transient increases of the EOD rate (E-scans) were detected from the z-scored first derivative of the 136 EOD frequencies. Variance and mean for the z-transform were based on pooled data of a given train- 137 ing trial. Accelerations exceeding a z-value of 1.5 were defined as E-scans and their location was de- 138 fined by the position at which the E-scan began.
139 All statistical analyses were performed using MATLAB and PAST 3 (Paleontological statistics software 140 package for education and data analysis version 3.1). Normality of data was examined by the 141 Shapiro–Wilk test and tested for homogeneity of variance with Levene’s test (significance criterion of 142 p d 0.05 in both cases). The appropriate parametric (t-test for linear regressions) or non-parametric 143 tests (Wilcoxon-signed rank test, Mann-Whitney pairwise test and Kruskal-Wallis test) were used 144 accordingly and are indicated in the results section and captions throughout. Data used for multiple 145 comparisons was post-hoc corrected (Bonferroni) when necessary using p d 0.05 for significance. 146 Transition probabilities of SPMs were tested using Pearson’s χ2 test.
147 Behavioral classification. We classified the behavior based on clustering algorithms as previously 148 described (Braun et al., 2010; Geurten et al., 2010; Hofmann et al., 2014). Kinematics were clustered 149 hierarchically (Ward’s criterion) and the quality and stability of the clusters were assessed to deter- 150 mine the optimal number of clusters. Next, data was clustered applying a k-means algorithm with the 151 optimal number of clusters regarding quality and stability (ten in our case). The resulting centroid of 152 each cluster (thrust, slip and yaw velocities) was used to express the kinematic properties of identi- 153 fied clusters that we here refer to as “prototypical movements” (PMs). PMs resemble the basic mo- 154 tor components of the recorded behavior on a frame-by-frame basis. PMs with yaw velocities above 155 half of the maximum observed yaw velocity (64°· s-1) were defined as left or right turn PMs, respec- 156 tively. PMs where the thrust velocities exceeded 75% of the maximum observed thrust (43.5 cm· s-1) 157 were considered a high thrust PMs, while PMs with thrust values lower than 25% of this maximum 158 were considered low thrust PMs. Using this categorization, we calculated thrust- and turn-triggered 159 averages of electromotor behavior (E-scans) and the quality of the sensory input (Fisher information, 160 see below). From these averages we compared the E-scan probability and Fisher information for the 161 100 ms surrounding different PMs.
162 To characterize behavior on larger timescales, we calculated the transition probabilities between 163 PMs. The probabilities were processed in a hidden Markov Model to determine the most frequent
7
164 sequences of four consecutive PMs, termed “super-prototypical movements” (SPMs) (Braun et al., 165 2010). Their complexity was reduced by combining SPMs with comparable PM composition (e.g. 166 merging PM-sequence 1-1-1-2-1-1-2-2 with PM-sequence 2-2-1-2-2-2-1-1). The 300 most frequent 167 SPMs were then further analyzed with respect to where (arena) and when (learning stage) they oc- 168 curred. We then focussed on the SPMs that changed the strongest between learning stages I and III. 169 That is, in each area of the arena we obtained the SPMs with the strongest drop in their recruitment 170 probability between stage I against III as well as those that showed the strongest increase in their 171 recruitment probability when comparing learning stages III against I. SPMs were again categorized by 172 their yaw and thrust velocities. In turn-dominated SPMs the mean yaw exceeded at least half of the 173 maximum mean yaw velocity of all SPMs (29°· s-1). The remaining SPMs were grouped based on the 174 maximum average thrust velocity (42 cm· s-1), resulting in low (<25% of max), medium (between 25- 175 75% of max) and high thrust SPMs (>75 % of max). The spatial distribution of different SPMs was 176 analyzed based on the x and y coordinates of the first PM in each SPM sequence. This was accumu- 177 lated and fitted with a 2-D Gaussian. For visualization we shown the area of these fits where the SPM 178 probability is above 0.1%. On average 89% of the individual data falls within this contour (range: 77- 179 96%).
180 Electric image model. Electric images (EI) were computed with software developed by Rother 181 (Rother, 2003). This approach was verified and utilized in previous studies (Rother et al., 2003; 182 Migliaro et al., 2005; Sanguinetti-Scheck et al., 2011; Hofmann et al., 2013, 2017; Pedraja et al., 183 2014). The model consists of a geometric reconstruction of the fish´s body and a calculation of the 184 transcutaneous field by solving the Poisson equation for the fish’s boundary using the Boundary Ele- 185 ment Method. Briefly, this method determines the boundary electrical distributions solving a linear 186 system of M · N equations for M poles and N nodes, with the unknown variables being the trans- 187 epithelial current density and potential at each node (Pedraja et al., 2014). The trans-epithelial cur- 188 rent density and potential is calculated for each node and linearly interpolated for the triangles de- 189 fined by the nodes, forming the geometry of fish and objects (Fig. 1A). From this the electric images 190 for each trajectory were calculated as the difference between amplitude of positive EOD peak in 191 presence and absence of the object (Fig. 1B).
192 Fisher information analysis. To estimate how informative the electric images (sensory input) are with 193 respect to an estimate of the location of the object, we calculated the Fisher information. As detailed 194 elsewhere (Silverman et al., 2013; Miller et al., 2016) the formal definition of Fisher information 195 when estimating an unknown parameter is p( | ) 1 ( ) = ( | )
8
196 where ( | ) is the probability of observing given . In our case, the measurement depends on a 197 function Υ(∙) of the parameter , 198 = Υ( , ) + ,
199 where Υ( , ) is the electric current density at the nodes of the fish at the fish’s location ( ) and is 200 zero mean noise with variance . The Fisher information then is calculated as